Years ago, there was a price war between public clouds. Back in 2014, to pick one example, Amazon’s AWS cut its prices in response to Google’s recently launched competing service.
Since those heady days, the cloud infrastructure market has matured and changed. Sure, AWS is still top dog, with Microsoft and Google working to both snag share from the leader (and one another). But the era of seemingly endless price cuts has been overtaken by a different market narrative: While building on public cloud services is inexpensive to start, it can become far less so over time.
That Dropbox made the choice to build out its own infra remains an interesting, if isolated, data point. (TechCrunch’s coverage from the event back in 2017 is worth your time.) We wanted to get a better vibe for what founders and CEOs are thinking about their public cloud choices and the strengths and weaknesses thereof.
So we got a hold of a few companies that we’re tracking, collecting input from BuildBuddy (early stage, YC backed, delivering a managed service), Monte Carlo (midstage, high growth, data focused), and Egnyte (late stage, profitable, a near-IPO company with a cloud storage and productivity focus) to get a broad view.
We surveyed the three founders and included their full replies below. But first, a few observations on their answers.
Don’t build alone
The number of companies that have built on a public cloud and later went solo is slim. Despite Dropbox managing the transition, and later finding gross-margin leverage in the effort, most companies that build on public clouds stay there.
That appears set to remain the case. The two younger companies we surveyed mentioned the required scale to make such a transition economical. Egnyte’s CEO, the leader of a company that has a history of cloud storage — meaning that surely it has the required scale, right? — mentioned some more modest cases where it may use its own hardware instead of public cloud services. But if Egnyte is still content to use public cloud infra, well, we can presume that nearly every startup is going to stay put as well.
Mostly (cloud) monogamous
Both BuildyBuddy (GCP) and Monte Carlo (AWS) are single-cloud companies. Egnyte has some workloads on clouds that are not its main, but noted that it is somewhat concentrated. As before, we’re seeing similar answers from each company, size to the side. This is why AWS et al. work with startup accelerators; if you get a company aboard your public cloud when it’s young, you have (nearly) a customer for life.
It appears, per the founders’ responses, that at a certain global footprint, companies may use more localized offerings. But that appears to be more edge case than main thrust for yet-private companies. Once a tech company is spinning up geospecific efforts, it’s probably ready to go public, yeah? Mostly, instead, cloud monogamy is the generally accepted operating principle for startups.
This is bullish for long-term public-cloud growth, and perhaps evidence that there is a blank space in the market when it comes to helping software companies break from their cloud overlords and go solo — or more indie, at least. After all, who wants to pay a good chunk of their gross margin to Amazon, say, so that it can make more lackluster television shows? Or to Google to de-risk Alphabet’s aging ad empire? Or to Microsoft so that it can keep making us use Teams?
Today, however, there do not appear to be clear offramps for public cloud usage among private tech companies. That feels inefficient. Read the following for the why.
Siggi Simonarson, founder, BuildBuddy
Are you building your startup atop a single public cloud or using a multicloud setup?
Our cloud product is built on a single public cloud (GCP), but we have self-hosted customers that host on AWS, Azure and their own custom hardware. We also run Mac workloads on MacStadium.
If you aren’t multicloud today, will you be in a few years’ time? Why?
I suspect we’ll be running some workloads on AWS at some point in the future. The big pull here is from customers that run their deployments on AWS and don’t want to pay large network egress costs to pull data from our servers running on GCP to their servers running on AWS.
Which public cloud provider do you use the most?
GCP.
What figured into your decision to pick your current public cloud provider?
Being a team of ex-Googlers, we were more familiar with GCP and the “Google way” of doing things. We also find their GCP console to be much easier to use than Amazon’s, which can sometimes feel like it’s never been seen or touched by a UI designer. Also, Google’s native Kubernetes support with GKE (which we use for many of our workloads) is way better than Amazon’s EKS offering in our experience. [Editor’s note: GKE stands for Google Kubernetes Engine, while EKS stands for Elastic Kubernetes Service, essentially the Kubernetes-focused offerings from both Alphabet and Amazon’s public clouds.]
How happy are you with its pricing and pace of development?
Cloud costs are high — particularly network egress costs and SSDs [solid-state drives]. This sometimes makes us dream of running our own hardware, but it just doesn’t make economic sense yet.
Are there enough public clouds in the market today that you feel like there is healthy competition for your startup’s business?
Yeah, and there are some exciting new teams taking a run at the “big 3” like fly.io and Cloudflare.
Where do you find that public clouds limit or accelerate your product development?
Network egress costs and the lack of Mac machines on GCP.
What might change your mind regarding using public cloud infra and pulling a Dropbox and building your own capacity?
It all comes down to scale. Running your own hardware (or using a bare metal provider like Equinix or OVHcloud) means you need to be able to justify the operational cost of managing the machines — which means the savings have to be large enough to pay for the additional headcount, effort, focus, time, etc.
Monte Carlo, CEO, Barr Moses and CTO Lior Gavish
Are you building your startup atop a single public cloud or using a multicloud setup?
Single cloud — AWS.
If you aren’t multicloud today, will you be in a few years’ time? Why?
Unlikely. There’s a lot of efficiency to be had from focusing on a single cloud, and it’s unlikely the advantages of a multicloud approach could offset it.
Which public cloud provider do you use the most?
AWS, Snowflake.
What figured into your decision to pick your current public cloud provider?
Familiarity in the founding team and proven, battle-tested capabilities around all the core services we needed.
How happy are you with its pricing and pace of development?
Pretty happy.
Are there enough public clouds in the market today that you feel like there is healthy competition for your startup’s business?
Generally speaking, yes. However, switching costs are massive, so in practice competition is limited.
Where do you find that public clouds limit or accelerate your product development?
- Deploying on the public cloud has made it easy to adopt technologies and services without investing resources in deploying and maintaining them.
- Scaling services as we grow is relatively easy.
- New employees are familiar with the technologies we use (compared to the case where we would have run our own cloud).
- There’s good support and integration points with common tools we use (like Snowflake, Databricks, Datadog, etc.).
What might change your mind regarding using public cloud infra and pulling a Dropbox and building your own capacity?
Companies need to reach massive scale with substantial pressure on margins to consider that as a worthwhile investment.
Vineet Jain, CEO and founder, Egnyte
Are you building your startup atop a single public cloud or using a multicloud setup?
We are using a multicloud setup with the majority of our deployment on one cloud. We leverage multiple clouds to deliver better service in regions that are underserved by our primary provider. We can offer georesidency across all regions served by any of the mainstream providers, provide faster ingress from all regions and leverage services (such as ML) that are better offered by specific providers for specific industries (such as life sciences).
Which public cloud provider do you use the most?
GCP.
What figured into your decision to pick your current public cloud provider?
Google’s reputation for best-in-class networking, containerization, ML and scalable storage systems plus volume discounts.
How happy are you with its pricing and pace of development?
We have been able to negotiate reasonable discounts using pre-commits. Regarding the pace of development, GCP’s pace for introducing new services is slower than AWS and could be faster.
Are there enough public clouds in the market today that you feel like there is healthy competition for your startup’s business?
Yes.
Where do you find that public clouds limit or accelerate your product development?
Different IAM models, machine types and idiosyncratic models need to be accounted for to support multiple providers. For example, some object stores support streaming inserts and some don’t, and compute sources (for example GPUs) may be available in only specific regions. But, overall we do get the ability to elasticize our computing and focus on application development rather than base infrastructure by using managed services provided by the cloud vendors.
What might change your mind regarding using public cloud infra and pulling a Dropbox and building your own capacity?
We constantly evaluate build-versus-buy choices for the various infrastructure components. Cloud providers are just one, albeit large, option. As we scale we may use CDNs for selective traffic to offset egress costs, push ML to our endpoints (mobile, desktop, hybrid appliances) and deploy our own infrastructure for non-elastic computing such as centralized logging and monitoring.